Comment by walterbell
1 day ago
> There is no such thing as a common subset of Markdown.
That was true before the widespread use of generative AI. LLM-generated markdown could _become_ the most common subset of Markdown, since machines can generate Markdown faster than humans.
“LLM generated Markdown” is not a coherent description of a language. LLMs can generate anything they have seen in their training data, which includes many incompatible dialects of Markdown.
Major LLMs (e.g. ChatGPT) have a copy button for their output, yielding markdown that is already being consistently rendered by other apps, reflecting what was consistently rendered as HTML. Presumably LLM vendors have built a deterministic way of generating spec-compliant HTML and their well-defined dialect of Markdown, otherwise their chat UI output would not render consistently.
I have certainly seen LLMs generate broken Markdown, so it’s presumably a “hope it works” thing.